This section is for builders who want **read access** to chain data and **limited writes** from automated workflows. MCP servers, SDK modules, and explicit permission scopes support that pattern. They do not replace a threat model: you still decide what an agent may spend, to which addresses, and for how long. ## Core pillars | Pillar | Description | Primary tool | | :--- | :--- | :--- | | **Connectivity** | Expose chain reads and scoped actions to MCP clients. | [Rootstock MCP Server](https://github.com/rsksmart/rootstock-mcp-server) | | **SDK** | Read state and submit guarded transactions from your stack. | [Rootstock SDK base](https://github.com/rsksmart/sdk-base) | | **Indexing** | Historical and aggregate data for dashboards and automation. | [The Graph on Rootstock](/dev-tools/data/thegraph/) | ## Key concepts ### 1. Model Context Protocol (MCP) MCP connects clients to tools and data sources. The **Rootstock MCP Server** can read balances, contract state, and gas estimates so responses reflect current chain data. Writes still need user-approved keys or policies you control. ### 2. Spend permissions and session keys Automation should not hold an unrestricted private key. Use spend permissions or session scopes that limit amount, destination, and duration for any transfer a model can request. ### 3. Rules-based actions With clear rules and permissions, a workflow can react to on-chain signals (for example vault or price conditions). You own the policy, monitoring, and rollback path. No guide here guarantees returns or strategy performance. ## Tools * **Rootstock MCP Server:** Exposes chain reads and safe actions to MCP clients. * **LangChain / Eliza:** Orchestration frameworks you can point at Rootstock RPC and wallets. * **Safe SDK:** Multi-sig and policy-based execution for high-value automation. ## Implementation guides Published guides in this section are listed below.